Perbandingan Algoritma Naïve Bayes, SVM Dan Xgboost Dalam Klasifikasi Teks Sentimen Masyarakat Terhadap Produk Lokal Di Indonesia

Abstract

Marketplace has become a popular online transaction medium with various features taken, one of the features that can be used for research is online reviews. Reviews can also be used as a data source for making various management decisions. Online reviews are very important in supporting purchasing decisions because of the development of e-commerce, there are more and more fake reviews so that more consumers are worried about online shopping. This cannot be denied because customer reviews can determine the level of customer satisfaction with the products that have been purchased. Sentiment analysis can be applied to Marketplace product reviews so that it can be used as product improvement suggestions for sellers and competitors so that they can find out what products are pleasing and needed by the community. Based on research that has been done, that the combination of Word2vec + XGBoost produces a higher F1 score of 0.941 followed by TF-IDF + XGBoost 0.940. Meanwhile, the SVM algorithm using vector space TF-IDF and Word2vec only produces 0.938 and 0.939. Meanwhile, Naïve Bayes has an F1-Score of 0.915 with TF-IDF and 0.900 with word2vec. Classification with word2vec in representing words into vectors is better than TF-IDF, this is because the advantages of word2vec are able to process semantic relations between words.

Sentiment Analysis of Public Opinions Towards Telkom University Post Pandemic

“…The number of Twitter users significantly increased to reach 10,645,000 users in 2020, which was the year the pandemic emerged. It then continued to rise, reaching 18.45 million users in 2022 [3].…”

Gender Classification of Twitter Users Using Convolutional Neural Network

“…For gender classification using N-gram feature extraction and Support Vector Machine classification algorithm with an accuracy of 93.2% with data totaling 65063 users where the data is in the form of profile picture, username, screen name, description, followers, and following, tweet, retweet, and favorite [3]. For comparing feature extraction between tf-idf and word2vec, using the XGBoost algorithm obtained an accuracy of 89.2% with tf-idf and 89.3% with word2vec [9]. Based on the above research, The difference between this research and previous research is gender classification by combining tweet data and descriptions in the Twitter user profile using word2vec and CNN for classification.…”